Hmmm.... I will look into this one thank you! 
 
*Edit*
 
OK... Running now... Looks like the 680's may be pretty fast with these as well...
 
 

Afterburner

Hmmm.... I will look into this one thank you! 

*Edit*

OK... Running now... Looks like the 680's may be pretty fast with these as well...

What's your usage?  here's the app_info.xml I'm using....  ave_ncpus and max_ncpus are the threads per task, without the app_info, POEM defaults it to 1, at least for one gpu.  Running values of 1 and a count of 0.5, my temps are higher than normal, running at values of 1.5, they are about normal, I would assume values less than 1 would further increase temps.  But I'm now using 3 threads for 2 gpu tasks and 4 threads for cpu tasks....  Still tweaking some things....
 
<app_info>
<app>
<name>poemcl</name>
<user_friendly_name>POEM++ OpenCL version</user_friendly_name>
</app>
<file_info>
<name>poemcl_1.3_windows_intelx86__opencl_nvidia_100</name>
<executable/>
</file_info>
<app_version>
<app_name>poemcl</app_name>
<version_num>103</version_num>
<avg_ncpus>1.5</avg_ncpus>
<max_ncpus>1.5</max_ncpus>
<platform>windows_intelx86</platform>
<plan_class>opencl_nvidia_100</plan_class>
<file_ref>
<file_name>poemcl_1.3_windows_intelx86__opencl_nvidia_100</file_name>
<main_program/>
</file_ref>
<coproc>
<type>NVIDIA</type>
<count>0.5</count>
</coproc>
</app_version>
</app_info>
 
 
Edit:  Taking it back down to 1 for ave and max_ncpus....  I thought I saw my cpu start downclocking at 60, but it doesn't appear so....  gonna push some more and see where I can get it....  Might be able to increase the oc some, too....

Afterburner

Showed 92% usage on one GPU (Device 0). Already done. Took 1800 seconds... 2925.23 points...

really?  I'm running them around 1400 each with two running on my 460....
 
In other news, up from 4.3 ghz to 4.4ghz and about 20 minutes stable crunching and only an increase of 2C from before the increase....  Pretty sure the voltage didn't really need to go up from 1.225 to 1.24, but I don't really have time right now to mess around finding the minimum voltage.

I got BOINC 7.0.28 built and running for Linux. However, the Poem OpenCL tasks keep restarting over and over with no indication as to what the issue is. I then upgraded to the latest version, 7.0.31 which did not help. I was able to successfully run Poem on my 8800GT in Linux using the same driver and BOINC version.
 
In Windows via my 580 and a single task running, I am seeing processing time of 858 seconds at 925 MHz and 835 seconds at 1 GHz. For some reason, when I try and run two or three tasks at once, my GPU usage does not go up. With a single task, usage is 45-54%. However, with two or three tasks, usage remains the same. I am thinking something driver related but otherwise I am not sure at this point.

it could be a scheduling thing inside BOINC....  Witht he app_info file, it started with an estimation of 40+ hours a task, after 14 hours, it has dropped to an estimation of 16 hours, so they're being ran high priority....
 
One other thing I noticed since moving it to the 3770/460, I can see it being more dependent on the cpu than other gpu projects.  It doens't run at a constant gpu usage, it dips every few seconds and then goes back to about 80%.  Through some tinkering, I noticed, the higher the percentage of cpu time you let boinc use, the less often Poem and gpu usage dips.  what is your cpu time percentage set to?  at 95%, it dips for a second about every 20 seconds, and at 80% it dips for a second about every 5 seconds.  I would assume, without much to really back it up, that at below 50%, it starts to cause problems, at 50% it should be one second running, one second paused, one second running, one second paused and once it starts getting lower than 50%, maybe it sees the longer than a second pause as "waiting"
 
How many tasks are you running per gpu?  it could potentially be a memory usage issue, as well, but precision shows my usage at 259mb with two tasks running.... 
 
Gonna tweak my app_info again, looking through the forums, the one who seems the be the app_info file "expert" on Poem is running 5 gpu tasks on one gpu using a total of 2.5 threads as opposed to my 2 tasks and 3 threads....  I doubt it'll be worth it for me and running 5, maybe three, but I'll see how things go running less than a thread each....
 
Nope, it is worse on usage jumping with less than a core, so I'm running 3 tasks using one thread each, I expect an increase in run time and as long as it is less than 2100 seconds average per task, it is more beneficial to run 3 task than 2.  Going from 1 task to 2, I saw an increase from about 1000 seconds to about 1400 seconds, rough slightly inflated averages....
 
Gonna have to give it more time, I have some between 1700 and 1900 and another at 2700....
 
Edit:  I gotta take a break from all of this....  Have a Statistics final, a Java web project and a Game project due in 48 hours....  Not too worried about the game and the final, the game is mostly finished, there are a few requirements that I just have to copy over from the last one, I didn't include them throughout the development because I knew the worked and didn't need to spend time worrying about them and even without doing the final, I pass the class with a high C....  I'll keep with updates when I need to take breaks....

I was finally able to get my 680 to run in Linux. It turns out the application restarting continuously was because I had the 9800 GT installed in the system system. Even though BOINC was set to ignore that GPU, for some reason the GPU had an effect over the OpenCL application. I tested with a single 680 standalone and the tasks processed successfully. These 680 cards haul with this project in Linux. The two tasks I ran last night completed between 655-660 seconds. However, like with my Windows system, if I put an app_info.xml file in place and run two tasks per GPU, I am not seeing any gains. The processing time doubles for each task. I have not quite figured that out yet.

This info is given as ref, as it is for amd HD 7770 gpus, it may not be relevant, but maybe it will give some help for tuning your nvidia setups??
 
SW:  Boinc v.7.0.25 64bit, amd v.12.4 driver, win7 pro 64bit, msi afterburner v2.2.1
HW:  C2Q qx6700@3.0GHZ, p6n diamond mb, 8gb ddr2 @ 800MHz, 2 x gigabyte HD 7770 gpu
 
5/26 1 wu/gpu 20&40GPU%  1.5hrs/wu  111,159ppd 
5/29 2 wu/gpu 20&34%  2.1hrs/wu  137,486
5/28 3 wu/gpu 20&35%   3.5hrs/wu   111,159ppd
 
NOTES:
1)  I read online that the windows client was very inefficient.  I havent tried linux yet, but linux mint 13 64bit is my next trial.
2)  I read online that faster mem improved performance, but saw no improvement when running the same gpus on another  motherboard(with C2Q Xeon at 3.0GHZ) using 8 GB of 1600MHZ ddr3.  I also got several failed wu, that may have been a setting issue.
3)  I noted that the 2nd gpu always started at 2d clocks, if no monitor was plugged in.  If a monitor was plugged in while crunching, it stayed at 3d clocks after unplugging the monitor.  Also worked with a kvm switching back and forth.
4)  I could never get any improvement in gpu useage %(max was 40% see above data), read in afterburner.  However, it would bounce much more(back and forth from the normal max to 0) if I did not keep 1 thread assigned/available per gpu wu.
 
good luck!  I hope to resume poem soon, after finishing some wcg work before they run out.

After several weeks of running Poem@Home, I like the performance vs load and temp of this project. Running a single task, my GPU temp is at 45C currently with 60% fan speed and power draw is quite a bit less than other projects. I still can't seem to get any additional gains from running 2-4 tasks at once. In both Windows and Linux, two tasks runs nearly double the time of one task and three tasks runs nearly triple the time of one task. Maximum GPU load appears to be 45-50%. However, runtime is low enough with a single task that I am not too worried about the scaling of running additional tasks. Also, there continues to be an issue with running this project with more than one NVIDIA GPU. Other users via the Poem forums appear to be having similar issues.